Question: Explain, using the definition of Recall, why relying on Recall alone would not give a complete picture of how well a cat classifier is performing on the dev/test set.

Sample answer: Recall only measures the percentage of all actual cat images that the classifier correctly labeled as a cat. It says nothing about how many non-cat images the classifier incorrectly labeled as cats. A classifier could achieve very high Recall simply by labeling almost every image as a cat, catching nearly all true cat images, but this would also produce many false positives on non-cat images. Because Recall focuses only on true cat images and ignores what happens to non-cat images, it cannot alone reveal whether the classifier is overall accurate or reliable; it must be considered alongside other information about the classifier's predictions.

Key points:

• Recall = percentage of all actual cat images correctly labeled as cat
• Recall does not account for non-cat images incorrectly labeled as cats
• A classifier could achieve high Recall by over-predicting the cat label
• Recall alone does not capture overall classifier reliability

Rubric: Full credit: correctly states Recall's definition, explains it ignores false positives/non-cat images, and explains a classifier could inflate Recall by over-predicting cats. Partial credit: states the definition but does not explain the limitation clearly. No credit: incorrect definition or no explanation of the limitation.